Google ships 3 new Gemini models. Just not the one everyone’s waiting for.

Google on Tuesday launched three new Gemini models: Gemini 3.6 Flash, a cheaper and faster 3.5 Flash-Lite, and 3.5 Flash Cyber, a model optimized for cybersecurity use cases that generally outperforms Anthropic’s 4.6 Opus model.

What is sorely missing in this lineup, and now weeks past its promised launch date, is 3.5 Pro, Google’s latest flagship model. Google first announced 3.5 Pro at its I/O conference in May with the promise of launching it in June. Google says the Pro model is “currently in testing with partners” and that it plans to make it broadly available “as soon as it’s ready.”

Never one not to look ahead, Google also says that it has started its pre-training run for Gemini 4.

3.6 Flash

Most of us didn’t expect to see an update to the Gemini Flash model before the Gemini 3.5 Pro model, but here we are.

The good news is that 3.6 Flash is a meaningful update to its predecessor — and it’s a bit cheaper. The input price per million tokens remains at $1.50, but Google reduces the price per million output tokens from $9 to $7.50.

Most of us didn’t expect to see an update to the Gemini Flash model before the Gemini 3.5 Pro model, but here we are.

Google also specifically notes that 3.6 Flash takes fewer reasoning steps and tool calls to achieve its goals in agentic workflows, which should also make it more cost-effective to run. According to Artificial Analysis, it uses 17% fewer tokens.

3.6 Flash Benchmarks

In most benchmarks, 3.6 Flash easily surpasses its predecessor, especially when it comes to coding. On the new DeepSWE software engineering benchmark, 3.6 Flash scores 49% vs 3.5 Flash ‘s 37%, for example. That’s not the greatest result, though. Anthropic’s Claude Sonnet 5 gets to 54% here, with the top-end frontier models scoring over 70%.

The team also saw large gains in the MLE Bench machine learning research benchmark (63.9% vs 49.7%) and the OSWorld-Verified computer use test (83% vs. 78.4%). Most modern models score in a similar range here,…